HOLMImpulse: Measuring Frequency- & Impulse-Response - Page 19 - diyAudio
Go Back   Home > Forums > Design & Build > Software Tools

Software Tools SPICE, PCB CAD, speaker design and measurement software, calculators

Please consider donating to help us continue to serve you.

Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
Reply
 
Thread Tools Search this Thread
Old 22nd June 2009, 11:49 PM   #181
KSTR is offline KSTR  Germany
diyAudio Member
 
KSTR's Avatar
 
Join Date: Jul 2007
Location: Central Berlin, Germany
Sorry Ask, there are still issues, it seems.

The fade-in is exponential, not raised-cosine, this gives a kinky spectrum and the default of 2 times the initial period time is problematic, too. With a start freq of 1Hz this means the stimulus is invalid during the first two seconds. Depending on the length used it may fade-in longer than the whole waveform runs. With shorter fade-in's the problem remains, there is a kink visible in the waveform itself with disastrous effects on the spectrum (and that would give artifacts in the distortion components, wouldn't it?)

Also the inverse still doesn't look right to me. I recognize that you construct the inverse in the freq domain so as to give a dirac by design: after reversing the inverse I get a reasonably (<> fully) perfect dirac when I convolve with the stimulus, no matter how corrupted it looks -- but only if I *don't* extend the stimulus on the ends with enough zeros, otherwise there are strong artifacts (echo pulse). This resulting "loopback" pulse should be artifact-free at any rate and should IMHO resemble the bandpass response used for the stimulus. That's at least what I did in my own experiment and I get a perfect artifact-free IR when loopback folding and the sprectra also are as clean as can be, so I doubt it's a computation artifact from Audition's convolver.


- Klaus
Attached Images
File Type: gif stimfadein.gif (18.2 KB, 470 views)
  Reply With Quote
Old 23rd June 2009, 07:05 AM   #182
wxa666 is offline wxa666  Germany
Account Disabled
 
Join Date: Jun 2009
Default Re: Re: Re: Re: Re: Re: Keep it simple!

Quote:
Originally posted by dlr


Your attitude has been from the beginning dismissive on that aspect, repeatedly as well. There may not be agreement on who is "arguing this thread to trash". Agreed that it's up to the "implementer". That is not you. I see no problem in discussing it. Start another thread if it bothers you.

Dave
Hi,

I didn't want to offend You, and yes I was dimissive right from the start. What would You tell me if I offer You - to bettern the sound - to stick Your fingers in Your ears? I doubt whether You would discuss that topic with me! John K. has had opened a thread on simulation integration some time ago. We could discuss that there. I have exercised dimissive argumentaton even before this thread ;-)

cheers
  Reply With Quote
Old 23rd June 2009, 08:44 AM   #183
diyAudio Member
 
Join Date: Mar 2003
Location: Denmark, Copenhagen
Default Logsweep

Thank you Klaus for your cooperation

The logsweep convolution is the core of HOLMImpulse, so we must agree on how this is working.
(To others: That is why Klaus and I are spending all this time not measuring speakers...)

Klaus and others I have got some questions...

1. Wave file bug?
You got error-messages when importing my wave-files. Do you still have these problems?

2. Fadein and fadeout amplitude
I use a 1/2 raised cosine to damp the amplitude of the signal.
The problem with the wavepacket is to get started.
Why do you want to use a full raised cosine?

3. Fadein and fadeout adjustments
My wavelengths approach is based on wavepacket ideas. If you want to make a 1Hz wavepacket, containing mostly 1 Hz frequencies, then you would need several seconds. I must of course improve it so I take into account that the frequency increases with a logsweep, but at 10Hz, I did not have this problem.
(You can set the fadin wavelengths to 0.1 if you prefer)

4. LogSweep and starting frequency
Let us say we measure at 44.1 kHz, from 1 Hz to 22 kHz, then 1/3 of the measurement time we are measuring frequencies below 20 Hz - We do not want that. BTW Farina starts at 20Hz - I find a good compromise starting at 10 Hz. See attachment for the 1Hz example

5. The inverse signal - Convolution is circular - zero extension
You must make a circular convolution of the inverse signal and the signal. Extending with zeroes? If you extend with zeroes before and after the signal and the inverse before making the circular convolution - yes then of course you will get garbage.

If extending the signal with zeroes before and after we must recalculate the inverse (HOLMImpulse does this when measuring)
I could add an option with adding zeroes to the saved signal and the inverse. Eg. if we have signal length 2^19 = 524288, then I will add 2^18 = 262144 zeroes before an after so that the new length of the zero extended signal and the NEW recalculated inverse is 2^20 = 2^19 + 2*2^18 = 1048576.
Is this what you want?
Attached Images
File Type: png screenshot020.png (27.7 KB, 449 views)
  Reply With Quote
Old 23rd June 2009, 11:46 AM   #184
JohnPM is offline JohnPM  United Kingdom
diyAudio Member
 
Join Date: Oct 2008
Default Re: Logsweep

Quote:
2. Fadein and fadeout amplitude
I use a 1/2 raised cosine to damp the amplitude of the signal.
The problem with the wavepacket is to get started.
Why do you want to use a full raised cosine?

3. Fadein and fadeout adjustments
My wavelengths approach is based on wavepacket ideas. If you want to make a 1Hz wavepacket, containing mostly 1 Hz frequencies, then you would need several seconds. I must of course improve it so I take into account that the frequency increases with a logsweep, but at 10Hz, I did not have this problem.
If you use a half cosine fade in the slope discontinuity at the end of the fade in will mean you deviate further than necessary from a truly log sweep which will affect the ability to properly separate the distortion harmonics. Using a fixed time period for the fade in (e.g. 100ms) works well.
  Reply With Quote
Old 23rd June 2009, 12:03 PM   #185
dlr is offline dlr  United States
diyAudio Member
 
Join Date: Mar 2005
Location: Canton, MA
Default Re: Re: Re: Re: Re: Re: Re: Keep it simple!

Quote:
Originally posted by wxa666


Hi,

I didn't want to offend You, and yes I was dimissive right from the start. What would You tell me if I offer You - to bettern the sound - to stick Your fingers in Your ears? I doubt whether You would discuss that topic with me! John K. has had opened a thread on simulation integration some time ago. We could discuss that there. I have exercised dimissive argumentaton even before this thread ;-)

cheers
Your arguments are incomplete and do not address relevant issues. The dismissive position is a disservice and does nothing to help "better the sound", quite the contrary.

Dave
  Reply With Quote
Old 23rd June 2009, 12:22 PM   #186
KSTR is offline KSTR  Germany
diyAudio Member
 
KSTR's Avatar
 
Join Date: Jul 2007
Location: Central Berlin, Germany
Default Re: Re: Logsweep

Quote:
Originally posted by JohnPM
If you use a half cosine fade in the slope discontinuity at the end of the fade in will mean you deviate further than necessary from a truly log sweep which will affect the ability to properly separate the distortion harmonics. Using a fixed time period for the fade in (e.g. 100ms) works well.
I think both my reasoning and my measurements back this.

I measured six combinations: Loopback, 48kHz, ASIO, start at 10Hz varying signal lengths (2^18 and 2^20) and the fade-in's and -out's :
a) fade-in 0, fade-out 0
b) fade-in 1, fade-out 30
c) fade-in 10, fade-out 300

The effect on THD can clearly be seen, on both ends. Note that for N=18 the fade-in=1 gives the best result (but still not the known level for the card, which is below -90dB at any from 10Hz to 1kHz). This might come from two things :
1) the kink in the waveform (in case of fade-in=0 right at the start) givensmore than necessary harmonics,
2) to a presumably way lesser extent, the loss of S/N as the stimulus has low amplitude in the fade areas.

- Klaus
Attached Images
File Type: gif fade_in_probs.gif (30.1 KB, 442 views)
  Reply With Quote
Old 23rd June 2009, 12:55 PM   #187
KSTR is offline KSTR  Germany
diyAudio Member
 
KSTR's Avatar
 
Join Date: Jul 2007
Location: Central Berlin, Germany
Hi Ask,
See my comments in italics after >> markers

Quote:
Originally posted by askbojesen
1. Wave file bug?
You got error-messages when importing my wave-files. Do you still have these problems?
>> No, this seems fixed.


2. Fadein and fadeout amplitude
I use a 1/2 raised cosine to damp the amplitude of the signal.
The problem with the wavepacket is to get started.
Why do you want to use a full raised cosine?
>> See my previous and John's post


3. Fadein and fadeout adjustments
My wavelengths approach is based on wavepacket ideas. If you want to make a 1Hz wavepacket, containing mostly 1 Hz frequencies, then you would need several seconds. I must of course improve it so I take into account that the frequency increases with a logsweep, but at 10Hz, I did not have this problem.
(You can set the fadin wavelengths to 0.1 if you prefer)
>> I'd prefer a adjustable absolute time (default ~100ms for both).


4. LogSweep and starting frequency
Let us say we measure at 44.1 kHz, from 1 Hz to 22 kHz, then 1/3 of the measurement time we are measuring frequencies below 20 Hz - We do not want that. BTW Farina starts at 20Hz - I find a good compromise starting at 10 Hz. See attachment for the 1Hz example
>> Not a big deal, only that the user must be aware of this. I agree 10Hz is a reasonable value for any real driver/speaker measurement.


5. The inverse signal - Convolution is circular - zero extension
You must make a circular convolution of the inverse signal and the signal. Extending with zeroes? If you extend with zeroes before and after the signal and the inverse before making the circular convolution - yes then of course you will get garbage.
>> Ah , this is the important thing. Alas I'dont know how to do circular convolution (never had a need or reason to do so, as of yet). However, I read Farina such as an aperiodic convolution shall be used. Except for calculation where the harmonics pulse are offset in time in the response, any zero padding of stimuls and/or convolution kernel beyond their relevant content shall not make any difference. As it is the case with my own set of data, where I see that the harmonic spacing follows exactly the formula given by Farina, using the length of the actual relevant content without any leading/trailing zeros.


If extending the signal with zeroes before and after we must recalculate the inverse (HOLMImpulse does this when measuring)
I could add an option with adding zeroes to the saved signal and the inverse. Eg. if we have signal length 2^19 = 524288, then I will add 2^18 = 262144 zeroes before an after so that the new length of the zero extended signal and the NEW recalculated inverse is 2^20 = 2^19 + 2*2^18 = 1048576.
Is this what you want?
>> not precisely, see above. I'd like to have data that is compatible whith standard convolving. But there is no need to add this extra option when I'm the only one requesting this as it were only for reference/educational purposes anyway.

One main point remains unanswered, that is that the "measurement signal IR/FR" in your graphs etc should contain the true response of the signal, except for one useful case: HF-attenuation for tweeter protecting, which should be a transparent pre-/post-EQ process. Together with this a stop frequency should be an additional option and that overall bandpass response should be visible, not EQ'ed out like it is now (and giving false data or interpretation of it).

- Klaus
  Reply With Quote
Old 23rd June 2009, 01:45 PM   #188
KSTR is offline KSTR  Germany
diyAudio Member
 
KSTR's Avatar
 
Join Date: Jul 2007
Location: Central Berlin, Germany
>> The logsweep convolution is the core of HOLMImpulse, so we must agree on how this is working.
>> (To others: That is why Klaus and I are spending all this time not measuring speakers...)

I like to add thist to my previous statements:
From a practical view most of this discussion could be considered to be academic (but it helps me a lot to see things clearer from this POV), while the goal will rather be to find a good general software setup and default settings to start with, for everyone to get reasonable results with real speaker measurements.

This software is very valuable, that is completely without doubt!

- Klaus
  Reply With Quote
Old 23rd June 2009, 01:55 PM   #189
gedlee is offline gedlee  United States
diyAudio Member
 
Join Date: Dec 2004
Location: Novi, Michigan
Quote:
Originally posted by KSTR
[B
This software is very valuable, that is completely without doubt!

- Klaus [/B]
This is very true. I hope to try it out this week, but still don't know if it will run under Vista 64.

Good work Ask!
  Reply With Quote
Old 23rd June 2009, 02:55 PM   #190
Pano is offline Pano  United States
diyAudio Moderator
 
Pano's Avatar
 
Join Date: Oct 2004
Location: Milliways
Blog Entries: 4
Default Re: Re: Re: Re: Re: Re: Re: Keep it simple!

Quote:
Originally posted by wxa666
What would You tell me if I offer You - to bettern the sound - to stick Your fingers in Your ears?
I would tell you "It works!" Ever been to a rock concert where the SPL was so high the system was distorting badly? Stick your fingers in your ears. If the distortion goes away - as it often will - it's not the system that's distorting. It your ears. Scary, but true.

So I would say that many a technique has its uses, as long as you find the "right use."

(Oh, now I see that Mr. WXA has been disabled)
__________________
Take the Speaker Voltage Test!
  Reply With Quote

Reply


Hide this!Advertise here!
Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off


Similar Threads
Thread Thread Starter Forum Replies Last Post
Measuring Frequency Response MCPete Multi-Way 10 10th February 2009 12:36 AM
Measuring Frequency Response furly Planars & Exotics 1 17th March 2006 06:59 AM
Measuring the frequency response of a mic? mr_push_pull Multi-Way 5 30th August 2004 12:26 PM


New To Site? Need Help?

All times are GMT. The time now is 11:15 PM.


vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2014 DragonByte Technologies Ltd.
Copyright 1999-2014 diyAudio

Content Relevant URLs by vBSEO 3.3.2